2.1 Manipulation of the Dataset

The datasets retrieved for this project is from the EVDS system, between December 2016 and January 2021 monthly. 72 data points are achieved in return, the data is merged and manipulated in the playground provided by R.

In addition, the data is checked numerically before going into further analysis and it is realized that there is 3 missing data points in Foreign Income dataset, which are April, May and June 2020. These are the months which pandemic started and it is assumed that both tourism and data collection process is interrupted due to pandemic’s restrictions. However instead of replacing these cells with 0, a naive approach is used in this scenario and the data is filled by the values of past year’s same months. Effects of pandemic is not lost, since its reflection can be seen in the posterior months when we numerically take a look on the dataset .

setwd('/Users/larahos/Desktop')
ForeignIncome=read.xlsx('Foreign-Income.xlsx')
AccommodationInd=read.xlsx('AccommodationInd.xlsx')
DollarBuy=read.xlsx('DollarBuy.xlsx')

#no values are assigned to April, May, June 2020 data:so we will use a naive approach and implement 
#the past years values to these months
ForeignIncome[52:54, "INCOME"] <- ForeignIncome[40:42, "INCOME"]

#we merged all dataset by using R 
joined_dt1 <- merge(ForeignIncome, AccommodationInd, 
                   by = 'Date')
data <- merge(joined_dt1,DollarBuy, by = 'Date')
colnames(data) <- c("Date", "Income","Accommodation_Index", "Dollar_Exchange_Rate")

#we will change all dates to year-month format which will be useful when we create a time series object
data$Date <- as.yearmon(x = data$Date)

#we created a time series object and we will check the numerical summary of the data since time bheaves as an Index rather than an Input in our case
data_ts <- xts(x = data[-1],order.by = data$Date,frequency = 12)

2.2 Numerical Analysis of the Dataset

The dataset which will be examined through the research is firstly be analyzed in the numeric sense at the beginning. At the first step, the summary of the time series object is printed. It is seen that mean and median values of Accommodation Index and Exchange Rate is so close, which can be concluded that they are likely to show a symmetric distribution in the following steps in contrast to the behavior of Income. The minimum value of income is extremely small, which may be an outcome of pandemics effect. Also there are some signs of seasonal patterns in head and tails of the Income dataset, which is related to the summer tourism in Turkey.

Furthermore, it is seen that the increase rate gained a positive acceleration for the last months of 2021. This fast increment is probably not reflected on the price index and foreign income rate due to the natural delay, which starts by the increase of producer price indexes and lastly reflects on the consumer.

All of these relations will be seen in a more clear way during the visualization of the time series data.

summary(data_ts)

##      Index          Income       Accommodation_Index Dollar_Exchange_Rate
##  Min.   :2016   Min.   : 286.0   Min.   :297.3       Min.   : 2.835      
##  1st Qu.:2017   1st Qu.: 867.2   1st Qu.:311.7       1st Qu.: 3.660      
##  Median :2019   Median :1240.0   Median :391.0       Median : 5.533      
##  Mean   :2019   Mean   :1512.9   Mean   :399.6       Mean   : 5.499      
##  3rd Qu.:2020   3rd Qu.:2166.5   3rd Qu.:458.4       3rd Qu.: 6.876      
##  Max.   :2022   Max.   :3864.0   Max.   :657.0       Max.   :13.528

head(data_ts)

##          Income Accommodation_Index Dollar_Exchange_Rate
## Jan 2016    825              304.03             3.006950
## Feb 2016    699              303.23             2.940662
## Mar 2016    904              303.37             2.891739
## Apr 2016    844              303.54             2.834738
## May 2016   1233              302.88             2.926595
## Jun 2016   1241              297.31             2.916986

tail(data_ts)

##          Income Accommodation_Index Dollar_Exchange_Rate
## Jul 2021   2201              531.38             8.612941
## Aug 2021   3077              548.97             8.475714
## Sep 2021   2483              574.76             8.511882
## Oct 2021   2808              608.43             9.139945
## Nov 2021   1477              623.24            10.523264
## Dec 2021   1089              656.99            13.528496

2.3 Visualization of Data

#we will all datasets line plots to have a general idea about their behaviour
par(mfrow=c(1,3))
ggplot(data_ts,aes(x=Index))+
  geom_line(size=1,color="brown",aes(y=Income))+
  theme_ipsum()+ggtitle("1. Time Series of Foreign Tourism Income")

ggplot(data_ts,aes(x=Index))+
  geom_line(size=1,color="red",aes(y=Accommodation_Index))+
  theme_ipsum()+ggtitle("2. Time Series of Accommodation Price Index")

ggplot(data_ts,aes(x=Index))+
  geom_line(size=1,color="brown",aes(y=Dollar_Exchange_Rate))+
  theme_ipsum()+ggtitle("3. Time Series of Exchange Rate USD/TRY")

indexes<-c("Income","Accomodation", "Dollar" )
plot(zoo(data_ts), main="Line Plot for 3 Datasets ", xlab="Date (Monthly)",ylab=indexes)

plot_num(data[,-1], bins=10)

ggplot(data_ts, aes(x=Income)) +
  geom_histogram(aes(y=..density..), colour="navyblue", fill="lightskyblue", bins = 30)+ 
  geom_density(alpha=.2, fill="pink", colour="red")

ggplot(data_ts, aes(x=Accommodation_Index)) +
  geom_histogram(aes(y=..density..), colour="navyblue", fill="lightskyblue", bins = 30)+ 
  geom_density(alpha=.2, fill="pink", colour="red")

ggplot(data_ts, aes(x=Dollar_Exchange_Rate)) +
  geom_histogram(aes(y=..density..), colour="navyblue", fill="lightskyblue", bins = 30)+ 
  geom_density(alpha=.2, fill="pink", colour="red")

ggplot(data_ts, aes(x=Income)) +
  geom_histogram(aes(y=..density..), colour="navyblue", fill="lightskyblue", bins = 15)+ 
  geom_density(alpha=.2, fill="pink", colour="red")+
  facet_wrap(~month(Index), ncol=3)+
  labs(title = "Monthly Histograms of Foreign Tourist Income 2016-2021", 
       x = "Total Income (million USD)",
       y = "Density")

min_max_norm <- function(x) {
  (x - min(x)) / (max(x) - min(x))
}
normalized_data_ts <- data_ts
normalized_data_ts$Accommodation_Index <- min_max_norm(data_ts$Accommodation_Index)
normalized_data_ts$Income <- min_max_norm(data_ts$Income)
normalized_data_ts$Dollar_Exchange_Rate <- min_max_norm(data_ts$Dollar_Exchange_Rate)



#we will plot all datas in a single frame to see their behaviour based on their normalized values
ggplot(normalized_data_ts)+geom_line(aes(x=Index, y=Income, color="Foreign Tourist Income"))+
  geom_line(aes(x=Index, y=Accommodation_Index, color="Accommodation Price Index" ))+
  geom_line(aes(x=Index, y=Dollar_Exchange_Rate, color="Dollar Exchange Rate" ))+
  ggtitle("Comparision of Normalized Values")

new.ts <- ts(normalized_data_ts, freq = 12)
Trend_Comparison <- decompose(new.ts)$trend
plot(Trend_Comparison)

2.4 Correlation Analyses

In addition to the visual inspection of the time series, calculating correlation is an important step to understand how tight the relationship is between the attributes. As the table shows, Price Index and Exchange Rate has a high correlation value, which is almost 1. They demonstrate a perfect positive correlation type of a behvaiour and the regarding correlation value is ranked as highly significant with stars.

ggpairs(data[,c(2,3,4)])

ggcorr(data[,c(2,3,4)]) 

It is also useful to compute autocorrelation function when there exists seasonal data, until lag equals to 12. Small lags have a high correlation value, which is a sign of increasing trend in the data set. Also the ACF function is cyclical and that is a typical behavior of seasonal data.

acf(data_ts$Income, lag=12, main = "Foreign Visitors Travel Incomes Seasonal Analysis")

Since similar increase in trend was seen in visualization section, a further analysis will be conducted on Foreign Visitor Travel Income dataset. The negative effects of pandemic hit the tourism season harder than most of other datasets. In the figure below, the months after pandemic is removed from all datasets, to see the correlation in a more stable enviroment.

without_covid_data <- data[-(51:72),]
ggpairs(without_covid_data[,c(2,3,4)])

As observed in the figures above, the correlation value between Income and other variables are significantly increased, which is parallel to the pre-study research conducted and proves the relationship between dollar rate and tourism income. As the exchange rate gets higher, it becomes more cheaper for a foreign traveler to spend their holiday in Turkey and demand increases, which eventually results in the increase of income. However, as stated before, this is not the only factor affecting the tourism income.

2.5 A Supportive Analysis Based On Google Trends

Google Trends is a useful tool to analyze the demand based on the total search volume of a keyword. The hypothesis constructed in these study is that tourism demand is related to the dollar exchange rates. Therefore, “Turkey Flight” keyword for the same time period is analyzed. As the demand increases, the volume is supposed to increase. These relation will be analyzed based on the boxplot:

FlightSearch <- read.xlsx('GoogleTrend.xlsx')

FlightSearch$Date <- as.yearmon(x = FlightSearch$Date)


ggplot(FlightSearch, aes(x=Turkey.Flight))+
  geom_boxplot(color="darkgrey",fill="brown", alpha=.4) +
    facet_wrap(~year(Date), ncol=3)+
  theme_ipsum()+
  ggtitle("Distribution of 'Turkey Flight' Keyword Search")

It is seen that the median values of boxplots shows a non decreasing pattern through 2016-2019. After the effects of pandemic, it got worse but started to increase again.

Another keyword used to further analysis is “Turkey Hotel Price”. As seen below, it shows a similar pattern with the seasonal sturcture of the tourism income. However there is a remarkable point in this visualization: After January 2021, Turkish Lira started to devaluate faster than many other country, and there occured an unexpected in the search volume of the subject keyword, even though it is not the season for summer tourism. The sharp increase in exchange rate resulted in the increase in demand off-season, too. When 2021 November-December data is compared to the 2018 of the respective months, higher values are seen in 2021 than the peak year of tourism in Turkey. This is a subject that relation of causality can be searched.

HotelSearch <- read.xlsx('GoogleTrendsHotelPrice.xlsx')
HotelSearch$Date <- as.yearmon(x = HotelSearch$Date)
DollarBuy$Date <- as.yearmon(x=DollarBuy$Date)
joined_dt2 <- merge(DollarBuy, HotelSearch, 
                   by = 'Date')
normalized_joined_dt2 <- joined_dt2
normalized_joined_dt2$DOLLAR.BUY <- min_max_norm(normalized_joined_dt2$DOLLAR.BUY)
normalized_joined_dt2$Turkey.Hotel.Price <- min_max_norm(normalized_joined_dt2$Turkey.Hotel.Price)

ggplot(normalized_joined_dt2)+geom_line(aes(x=Date, y=DOLLAR.BUY, color="Dollar Exchange Rate"))+
  geom_line(aes(x=Date, y=Turkey.Hotel.Price, color="Hotel Price Search" ))